A Doctor in Every Phone: Anthropic and OpenEvidence Bring Free Clinical AI to 100 Countries

For years, the debate over medical AI in rich countries has centered on whether large language models should be allowed near high-stakes clinical decisions at all. Meanwhile, a quieter and arguably larger question has been maturing elsewhere: what happens when the same technology reaches places that have never had a subspecialist, a journal subscription, or reliable electricity — but where nearly every physician carries a smartphone?

This week, that question got a concrete answer. Artificial intelligence startup Anthropic and medical knowledge platform OpenEvidence announced a collaboration to bring AI-powered clinical decision support to physicians in regions that until now had no practical access to the technology. The specialized version of OpenEvidence’s platform will be free to healthcare providers in roughly 100 low- and middle-income countries (LMICs), the companies said on Tuesday, September 22. Anthropic is providing back-end model capacity, while OpenEvidence tailors the system to regional healthcare infrastructure. Financial terms were not disclosed.

What OpenEvidence Actually Does

Understanding why this matters requires understanding what OpenEvidence is — and, just as importantly, what it is not. Founded by Daniel Nadler, the platform is an AI copilot for physicians that answers clinical questions by drawing on peer-reviewed medical research and treatment guidelines, with citations attached to every claim. It is not a general-purpose chatbot with a medical prompt; it is a retrieval-grounded system built for the point of care, where a physician has minutes, not hours, to work through a differential diagnosis or check a drug interaction.

The usage numbers explain why Anthropic wanted this partnership. In August 2026 alone, U.S. clinicians consulted OpenEvidence 42 million times, according to Nadler. He estimates that in 2026, several hundred million Americans will have been treated by a doctor who used the platform to help guide care. The product is already free to clinicians in the United States and Europe. What was missing was everywhere else — precisely the places where limited access to medical literature, specialist expertise, and continuing medical education does the most damage to patient outcomes.

The scale of that gap is worth pausing on. The World Health Organization has long documented that the density of physicians and specialists in LMICs is a small fraction of high-income-country levels, and that clinical decisions in under-resourced settings are frequently made without access to current evidence. An internist at a referral hospital in a capital city may have institutional journal access; a district-hospital doctor in a rural province typically does not. The result is a kind of evidence deserts — care practiced decades behind the published frontier, not because clinicians lack skill, but because the knowledge infrastructure is absent.

Context-Adaptive, Not Just Translated

The most technically interesting part of the announcement is not the price tag (free) or the model (Claude). It is the explicit commitment that the LMIC version is being adapted, not exported. “One hundred percent of what we are building for these places is context adaptive,” Nadler told Reuters.

That distinction addresses the sharpest criticism of exporting clinical AI to low-resource settings: systems trained primarily on data from high-income countries may not reflect local disease patterns, diagnostic resources, or available treatments. A guideline that assumes ready access to an MRI machine, a broad formulary, or a specialty referral pathway is worse than useless in a clinic that has none of the three. OpenEvidence says it takes local healthcare infrastructure into account when tailoring answers — so the system should recommend what is actually achievable where the question is being asked.

This builds on groundwork the company laid earlier in the year, when it began working with health organizations in Rwanda and Botswana to adapt its tools for settings where disease patterns, diagnostic resources, and available treatments differ markedly from wealthy countries. The new initiative scales that pilot approach to a list of roughly 100 countries, including Uganda, Angola, Sudan, Haiti, and Mongolia, according to a list provided by OpenEvidence.

Even the delivery assumptions are adapted to the terrain. Many health facilities in the target regions lack around-the-clock electricity — but, as Nadler pointed out, most physicians do have smartphones. That flips the deployment model: instead of wiring up hospitals, the clinical knowledge layer rides on the device already in every doctor’s pocket. Access to quality medical information that a decade ago would only have been available at leading institutions like the Mayo Clinic becomes, in effect, a phone plan away.

Why Anthropic Is Doing This

For Anthropic, the move continues a strategy of positioning Claude as the AI layer for serious institutional work — healthcare, life sciences, government — rather than the consumer chatbot arena. The company had already launched Claude for Life Sciences, and in May 2026 it announced a partnership with the Gates Foundation focused heavily on improving health outcomes in low- and middle-income countries, home to roughly 4.6 billion people. The OpenEvidence collaboration is a natural extension: a concrete deployment that aligns with the company’s stated mission of seeing AI benefit humanity broadly, while putting Claude into the daily workflow of physicians at a scale few products achieve.

Anthropic president Daniela Amodei framed the deal as fixing a market failure rather than creating a market. “The technology itself has advanced so dramatically that the sort of structure needed for bringing OpenEvidence to low-resource regions is there,” she said, “but the market incentives alone would not let it happen without this type of entrepreneurial, philanthropically minded work.”

That framing is candid: none of these countries were going to be profitable enterprise customers on any near-term horizon. The deal exists because a frontier lab with spare model capacity and a medical-AI company with a proven product decided the reputational, mission, and long-term strategic value of global deployment outweighed the near-term revenue. Reuters also reported last week that Anthropic has built a lab to do physical biology work as it pushes its AI ambitions into drug science — the OpenEvidence deal extends the same healthcare push from the research end to the delivery end.

The Optimists and the Skeptics

Early clinical reaction from the target regions skews hopeful. Dr. Ahmed Bendary, a cardiologist at Benha University in Egypt, said expanding access outside the United States and Europe “would be a tremendous leap forward, particularly in regions where institutional subscriptions to major medical journals are limited, making evidence-based, point-of-care tools even more vital.”

The caveats are real, however, and the companies’ own framing acknowledges them. Critics of clinical AI deployment in LMICs caution that models can carry high-income-country biases in their training data, that point-of-care tools can be over-trusted by busy clinicians, and that no regulatory infrastructure exists in many target countries to audit such systems. Localization at the scale of 100 countries — each with different formularies, guidelines, disease burdens, and languages — is an enormous engineering and clinical challenge, and the proof will be in the details of each regional rollout, not the press release. Notably, the announcement did not specify which Claude models power the deployment, what evaluation or safety monitoring will accompany it, or how the “context adaptive” tailoring will be validated country by country.

There is also a question of durability. “Free at the point of use” is a promise that depends on two companies’ continued willingness to subsidize inference costs for an unbounded user base. Anthropic and OpenEvidence did not disclose the financial terms, how the costs are split, or how long the commitment runs.

What It Means

Strip away the caveats, and this is still one of the largest deliberate deployments of frontier AI in global health to date: roughly 100 countries, free access for physicians, and a product with demonstrated traction at 42 million monthly U.S. clinical queries. If the context-adaptive engineering holds up, it narrows the evidence gap between a teaching hospital in Boston and a district clinic in Gulu at a stroke — not by building new hospitals, but by reclassifying world-class medical knowledge as something that arrives over a phone network.

The collaboration also says something about where the AI industry’s competitive frontier is drifting. A year ago, labs competed on benchmark scores. This week, Anthropic shipped compute to clinics. As model capabilities increasingly commoditize, distribution into high-stakes professional workflows — and the public-good deployments that build trust in them — is becoming the differentiator. The physicians of Uganda, Mongolia, and Haiti are about to find out whether that trust is warranted.